TapStr: A Tap and Stroke Reduced-Qwerty for SmartphonesMohammad Akbor Sharif

Human-Computer Interaction GroupUniversity of California, MercedMerced, California, United States

msharif@ucmerced.edu

Gulnar RakhmetullaHuman-Computer Interaction GroupUniversity of California, MercedMerced, California, United Statesgrakhmetulla@ucmerced.edu

Ahmed Sabbir ArifHuman-Computer Interaction GroupUniversity of California, MercedMerced, California, United States

asarif@ucmerced.edu

Figure 1: TapStr is an unambiguous reduced-Qwerty that maps the standardQwerty in a single row divided into four zones.

Tapping on a zone enters the central character and stroking on a zone towards one of the eight characters around the border

enters the respective character. Dwelling for 500 ms upon a tap or stroke enters the uppercase letters or the secondary charac-

ters, distinguished using a smaller font (right). Dwelling for 1000 ms activates or deactivates Caps Lock. Stroking towards ‘↙’

switches the layout for numeric and special characters (right) and stroking towards ‘,’ switches the layout for emojis.

ABSTRACT

TapStr is a single-row reduced-Qwerty that enables text entryon smartphones by performing taps and directional strokes. It isan unambiguous keyboard, thus does not rely on a statistical de-coder to function. It supports the entry of uppercase and lowercaseletters, numeric characters, symbols, and emojis. Its purpose is toprovide an economic alternative to virtual Qwerty when savingtouchscreen real-estate is more desired than the entry speed anda convenient alternative when users do not want to repeatedlyswitch between multiple layouts for mixed-case alphanumeric textand symbols (such as a password). In a short-term study, TapStryielded about 11 WPM with plain phrases and 8 WPM with phrasescontaining uppercase letters, numbers, and symbols.

CCS CONCEPTS

• Human-centered computing→ Text input; Gestural input.

KEYWORDS

Text entry; reduced keyboard; virtual keyboardACM Reference Format:

Mohammad Akbor Sharif, Gulnar Rakhmetulla, and Ahmed Sabbir Arif.2020. TapStr: A Tap and Stroke Reduced-Qwerty for Smartphones. InCompanion Proceedings of the 2020 Conference on Interactive Surfaces andSpaces (ISS ’20 Companion), November 8–11, 2020, Virtual Event, Portugal.ACM, New York, NY, USA, 4 pages. https://doi.org/10.1145/3380867.3426208

Virtual Qwerty has become the dominant method for text entryon smartphones. It is evident that with enough practice, mobileusers can reach a competitive entry speed with virtual Qwerty[12]. However, one problem with virtual Qwerty is that it occupies

Permission to make digital or hard copies of part or all of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for third-party components of this work must be honored.For all other uses, contact the owner/author(s).ISS ’20 Companion, November 8–11, 2020, Virtual Event, Portugal© 2020 Copyright held by the owner/author(s).ACM ISBN 978-1-4503-7526-9/20/11.https://doi.org/10.1145/3380867.3426208

about 40% of a smartphone display. Saving smartphone real-estateis crucial since cluttered display affects user performance [11]. Be-sides, the saved space could be used to display more of what hasbeen entered so far to keep the user aware of the context of a chatand the flow of a passage, which improves both writing speed [10]and quality [22]. Besides, entering special characters, symbols, andemojis is difficult with virtualQwerty (especially for novices) sinceit requires users to repeatedly switch between multiple layouts (typ-ically, four or more) for the intended characters. TapStr is a novelsingle-row 65×8 mm reduced-Qwerty that occupies only about 7%of a smartphone display. It uses only three layouts to enable the en-try of alphanumeric and a range of special characters and emojis byperforming taps and directional strokes on the keyboard. The moti-vation of TapStr is not to replace the conventional virtual Qwertybut to provide an economic alternative when saving touchscreenreal-state is more desired than the entry speed (for example, whenusers do not want to repeatedly scroll up and down to see the emailthey are responding to or the video they are commenting on) and aconvenient alternative when users do not want to switch back andforth between different layouts to enter mixed-case alphanumerictext and symbols (for example, when entering a password or URL).

1 TAPSTR KEYBOARD

TapStr maps the standard Qwerty layout in a single row dividedinto four zones. The zones are distinguished using different shedsof grey (Fig. 1). TapStr uses the Qwerty layout to facilitate skilltransfer [7, 17], but occupies only 5.24 cm2 of a 74.5 cm2 smartphonescreen, when a virtual Qwerty occupies 23–29 cm2, freeing up25–32% of the display. The first three 19.66×8 mm zones containnine characters each: one in the middle, four along the four sides,and four in the four corners. Users tap on a key to enter its centralcharacter and stroke towards the characters around the border toenter the respective characters (Fig. 1). A stroke can be initiatedanywhere on the zone containing the target character and endanywhere on the touchscreen, even outside the keyboard. TapStrregisters all touch-point movements over 10 pixels as strokes anddetermines its direction based on the start and end points. This

ISS ’20 Companion, November 8–11, 2020, Virtual Event, Portugal Sharif, Rakhmetulla and Arif

enables users to change the direction of a stroke midway when theymistakenly stroked towards an incorrect character. For uppercaseletters, users first tap or stroke for the intended letter, then dwellfor 500 ms. If a dwell is activated by mistake, users can slightlyshift the finger to disable it. The keyboard provides visual feedbackon dwell by flipping the layout from lowercase to uppercase. Thelayout returns to lowercase upon touch-up. Users can dwell for 1000ms to enable (or disable) Caps Lock. The fourth 6×8 mm zone isfor Enter, Backspace, and Space, which are entered by an up-stroke,tap, and down-stroke, respectively. Stroking towards ‘↙’ switchesthe layout for numeric and special characters and vice versa (Fig. 1).TapStr places all paired symbols side-by-side and groups similarsymbols together (e.g., places all mathematical operators on thethird zone) to facilitate mnemonic strategies. The layout categorizesthe most frequent symbols as primary characters and the leastfrequent ones as secondary characters1. Primary and secondarycharacters are visually distinguished in the layout using a bigger anda smaller font-size (Fig. 1, right), respectively. Symbols are enteredusing the same mechanism as letters: users tap or stoke to enterthe primary characters and dwell to enter the secondary characters.Dwelling provides visual feedback by flipping the font-size of theprimary and the secondary characters. The ‘,’ symbol switches thelayout for emojis. However, we disable this feature in the study. Wealso implemented a predictive system for the keyboard to enableword prediction and auto-correction. This feature was also disabledduring the study to eliminate a penitential confounding factor.

2 RELATEDWORK

There are several reduced keyboards available for mobile devices.Stick Keyboard [8] maps the four rows of Qwerty onto the homerow. Although designed for mobile devices, it was evaluated ona desktop computer with a physical prototype, where it yielded10.4 WPM. 1Line Keyboard [14] is a similar virtual keyboard fortablets. With the support of a statistical decoder, it yielded 30.7WPM in a longitudinal study. TenGO [21] maps Qwerty onto sixkeys and uses four extra keys for Space, Backspace, Shift, and tocycle through suggested words. This keyboard was not evaluatedin a user study. Gueorguieva et al. [9] designed a virtual keyboardto enable text entry using Morse code. In a longitudinal study, itreached about 7 WPM. Senorita [18] is a virtual chorded keyboardthat arranges all letters on eight keys laid out in a single row. Ina longitudinal study, it reached 14 WPM. Arif et al. [3] replacedthe Space, Backspace, Shift, and Enter keys of a virtual Qwertywith directional strokes to make room for numeric and specialcharacters in the main layout. In a study, it yielded 17.4 WPM.TaS [19] arranges all letters in a 4×2 grid alphabetically. Each keycontains one character in the middle and four characters alongthe four sides. Similarly, TapFlick [15] maps Qwerty onto threekeys in a single row. Each key contains nine characters: one inthe middle, four along the four sides, and the remaining four inthe four corners. Like TapStr, both TaS and TapFlick use taps anddirectional strokes for text entry. However, these keyboards occupyabout the same screen real-estate as conventional virtual Qwerty1Due to the absence of a frequency table for symbols, we studied the state-of-the-artvirtual keyboards and assumed that all symbols that are placed on a higher layout(which require fewer actions to access) are the most frequently used symbols and theones that are placed in lower layouts are the least frequently used symbols.

and do not fully support the entry of numeric and special characters.Besides, some of these keyboards reviewed here are ambiguous,thus rely on aggressive statistical decoders, which makes enteringout-of-vocabulary words very difficult, seldom impossible.

Recently, there has been a growing interest in miniature key-boards for smartwatches [1]. The most relevant to our work isSwipeKey [20] that arranges all letters of the English alphabet in a4×2 grid in an alphabetical order. Each key includes four charac-ters in the four sides, which are entered by performing directionalstrokes. However, this keyboard does not support the entry of nu-meric characters and special symbols.

Figure 2: The device and the app used in the study (left) and

two participants taking part in the study (right).

3 EXPERIMENT

We conducted a study to compare the performance of TapStr withGboard, the default Google Android keyboard [13].

3.1 Participants

Twelve participants aged 19–28 years (M = 21.5, SD = 3.1) tookpart in the study (Fig. 2). Four of them were female and eight weremale. They all were proficient in English and experienced virtualQwerty users (M = 8 years of experience, SD = 2.04). Ten of themwere right-handed and two were left-handed. They all received US$10 for volunteering.

3.2 Apparatus

The study used a Motorola Moto G5 Plus smartphone (150.2×74×7.7mm, 155 g) running on Android OS 7.0 Nougat at 1080×1920 pixels.All predictive features of the two keyboards were disabled to elimi-nate a potential confound. Text entry performance was recordedusing WebTEM2, a freely available web application [2].

3.3 Design

The study used a within-subjects design. The independent variableswere keyboard and phrase, and the dependent variables were theperformance metrics. We recorded the standard words per minute(WPM) and error rate (%) metrics. Words per minute is the averagenumber of words entered in oneminute, where a “word” ismeasuredas 5 characters [4]. Error rate is the average percentage (%) ofincorrect characters remained in the transcribed text. The study2WebTEM: A web application to record text entry metrics, https://WebTEM.site

TapStr: A Tap and Stroke Reduced-Qwerty for Smartphones ISS ’20 Companion, November 8–11, 2020, Virtual Event, Portugal

used two phrase sets. Plain phrases are from the MacKenzie andSoukoreff set [16] that are moderate in length (M = 29), contain afew uppercase letters but no numeric or special characters. Mixedphrases3 are from a different set [3] that are also moderate in length(M = 37) but contain on average 7% uppercase letters, 10% numericcharacters, and 7% symbols. An example phrase from this set is“$6.52 is way too much for a bottle of water!”. Hence, the design was:

12 participants ×2 keyboards (Qwerty and TapStr, counterbalanced) ×2 phrase sets (plain and mixed, counterbalanced) ×15 phrases = 720 phrases, in total.

3.4 Procedure

The study took place in a quiet room. Upon arrival, we demonstratedTapStr and explained the study procedure to all participants. Wethen collected their consents. We enabled participants to practicewith TapStr by entering free-form text for about five minutes. Wedid not ask them to practice with the Gboard since they all wereexperienced users of virtual Qwerty. We then started the mainstudy that required participants to transcribe fifteen phrases fromtwo different sets (plain and mixed) using two different keyboards(TapStr and Qwerty). The phrase sets and the keyboards werecounterbalanced to eliminate the effect of learning. In each condi-tion, random phrases were presented on the top of the screen, oneat a time (Fig. 2). We instructed participants to read the phrasescarefully, transcribe them as fast and accurate as possible, then pressEnter to see the next phrase. Error correction was encouraged butnot forced. Logging started after entering the first character andended with the last. We informed participants that they could takebreaks between the conditions or before they start typing a phrase.Upon completion of the study, they completed a short questionnairethat asked them to rate various aspects of the new keyboard.

4 RESULTS

For statistical tests, we removed all phrases that were missing morethan ten characters (9% of the data). We used a repeated-measuresANOVA for all analysis since the data did not violate the normalityor the sphericity assumptions.

4.1 Entry Speed

An ANOVA identified a significant effect of keyboard (F1,11 = 11.49,p < .01) and phrase (F1,11 = 13.40, p < .01) on entry speed. There wasno significant effect of block (F2,22 = 1.86, p = .18) but the keyboard× block (F2,22 = 4.00, p < .05) and the keyboard × phrase × block(F2,22 = 5.70, p < .05) interaction effects were significant. However,the keyboard × phrase (F1,11 = 0.65, p = .44) interaction effect wasnot significant. Fig. 3 illustrates average entry speed of the twokeyboards with plain and mixed phrases.

4.2 Error Rate

An ANOVA failed to identify a significant effect of keyboard (F1,11= 0.55, p = .48), phrase (F1,11 = 1.48, p = .25), or block (F2,22 = 0.87,p = .43) on error rate. The keyboard × block (F2,22 = 0.77, p = .48),the keyboard × phrase (F1,11 = 1.64, p = .25), and the keyboard ×

3Mixed phrases, https://www.asarif.com/pub/mixedset.txt

Figure 3: Average entry speed of the two keyboards with

plain and mixed phrases. Error bars represent ±1 standard

deviation (SD).

Figure 4: Average error rate (%) of the two keyboards with

plain and mixed phrases. Error bars represent ±1 standard

deviation (SD).

phrase × block (F2,22 = 0.00, p = .99) interaction effects were also notsignificant. Fig. 4 illustrates average error rate of the two keyboardswith plain and mixed phrases.

5 DISCUSSION

TapStr yielded a significantly slower entry speed (71.72%) thanQw-erty. We anticipated this since participants were new to the layout,thus needed time to learn it. Besides, TapStr requires performinglinear strokes to enter most characters, which most probably con-tributed towards the slower entry speed. Prior research showedthat stroke lengths reduce with practice, as users get more familiarwith a new layout [3]. Hence, it is possible that the keyboard willyield a much better entry speed in a longitudinal study. The 500 msdwell time for the uppercase letters and some special charactersmay have contributed towards the slower entry speed as well. Aprior work attempted to mitigate this by gradually reducing thedwell time as users get more familiar with the system [6]. Furtherinvestigation is needed to find out whether this approach benefitsTapStr. Entry speed with both keyboards were much slower for

ISS ’20 Companion, November 8–11, 2020, Virtual Event, Portugal Sharif, Rakhmetulla and Arif

mixed phrases compared to plain phrases. This is also expectedas entering numbers and symbols requires extra effort with bothkeyboards. But interestingly, entry speed with TapStr dropped at amuch lower rate (26%) compared toQwerty (41%), which indirectlysuggests that entering numbers and symbols with TapStr does notrequire as much effort as Qwerty. Significant interaction effects onentry speed suggest that entry speed during the study, particularlyfor mixed phrases, improved with practice. A Tukey-Kramer testconfirmed that entry speed for mixed phrases was much faster inthe last block compared to the first block (p < .05). Besides, withboth Qwerty (𝑅2 = 0.94) and TapStr (𝑅2 = 0.88), entry speed overthe blocks for mixed phrases correlated well to the power law ofpractice [5].

There was no significant main or interaction effects on errorrate. TapStr yielded a much higher error rate (2.69%) than Qwerty(1.05%). A deeper analysis revealed that text entry with TapStr be-came more accurate with practice (𝑅2 = 0.86), while no such trendswere observed with Qwerty. This is likely because participantswere already familiar with Qwerty, thus did not trigger as manyhit-and-miss as TapStr. Further, participants were becoming moreefficient in entering numbers and symbols with both keyboards(Qwerty: 𝑅2 = 0.85, TapStr: 𝑅2 = 0.99), requiring fewer correctiveactions, which correlate well to the power law of practice [5].

Questionnaire data revealed that participants were not very en-thusiastic about TapStr. Roughly half of them were against usingTapStr for both plain (42%, N = 5) and mixed phrases (50%, N =6), while the remaining were neutral about it. This demonstratesexperienced virtual Qwerty users’ reliance and confidence in thekeyboard they are familiar with. Further investigation is neededto find out if their initial impression of the keyboard changes afterusing it for a longer period of time. Participants were mostly neutralabout the learnability of the new keyboard. About 25% of them (N= 3) found the keyboard difficult to master, the remaining (N = 9)were neutral.

6 CONCLUSION AND FUTUREWORK

This paper presented TapStr, an unambiguous reduced-Qwertythat occupies only about 7% of a stock smartphone screen. It enablesthe entry of uppercase and lowercase letters, numeric characters,symbols, and emojis by taps and directional strokes. In a short-termstudy, TapStr yielded on average 11 WPM entry speed for plainphrases and 8 WPM for mixed phrases with only 2–3% error ratewithout the support of a statistical decoder or a predictive system.In the future, we will conduct a longitudinal study to investigateif the performance and user preference of the keyboard improvewith practice. We will also explore the potential of the keyboard onsmaller devices like smartwatches.

REFERENCES

[1] Ahmed Sabbir Arif and Ali Mazalek. 2016. A Survey of Text Entry Techniquesfor Smartwatches. In Human-Computer Interaction. Interaction Platforms andTechniques (Lecture Notes in Computer Science), Masaaki Kurosu (Ed.). SpringerInternational Publishing, Cham, 255–267. https://doi.org/10.1007/978-3-319-39516-6_24

[2] Ahmed Sabbir Arif and Ali Mazalek. 2016. WebTEM: AWeb Application to RecordText Entry Metrics. In Proceedings of the 2016 ACM International Conference onInteractive Surfaces and Spaces (ISS ’16). ACM, 415–420. https://doi.org/10.1145/2992154.2996791

[3] Ahmed Sabbir Arif, Michel Pahud, Ken Hinckley, and Bill Buxton. 2014. Exper-imental Study of Stroke Shortcuts for a Touchscreen Keyboard with Gesture-Redundant Keys Removed. In Proceedings of Graphics Interface 2014 (GI ’14).Canadian Information Processing Society, Toronto, ON, Canada, 43–50.

[4] Ahmed Sabbir Arif and Wolfgang Stuerzlinger. 2009. Analysis of Text EntryPerformance Metrics. In 2009 IEEE Toronto International Conference Science andTechnology for Humanity (TIC-STH). 100–105. https://doi.org/10.1109/TIC-STH.2009.5444533

[5] Stuart K. Card, Thomas P. Moran, and Allen Newell. 1983. The Psychology ofHuman-Computer Interaction. CRC Press.

[6] Steven J. Castellucci, I. Scott MacKenzie, Mudit Misra, Laxmi Pandey, andAhmed Sabbir Arif. 2019. TiltWriter: Design and Evaluation of a No-TouchTilt-Based Text Entry Method for Handheld Devices. In Proceedings of the 18thInternational Conference on Mobile and Ubiquitous Multimedia (MUM ’19). ACM,1–8. https://doi.org/10.1145/3365610.3365629 Article 7.

[7] Mikael Goldstein, Robert Book, Gunilla Alsiö, and Silvia Tessa. 1999. Non-Keyboard Qwerty Touch Typing: A Portable Input Interface for the Mobile User.In Proceedings of the SIGCHI conference on Human Factors in Computing Systems(CHI ’99). ACM, Pittsburgh, Pennsylvania, USA, 32–39. https://doi.org/10.1145/302979.302984

[8] Nathan Green, Jan Kruger, Chirag Faldu, and Robert St. Amant. 2004. A ReducedQwerty Keyboard for Mobile Text Entry. In CHI ’04 Extended Abstracts on HumanFactors in Computing Systems (CHI EA ’04). ACM, New York, NY, USA, 1429–1432.https://doi.org/10.1145/985921.986082

[9] Anna-Maria Gueorguieva, Gulnar Rakhmetulla, and Ahmed Sabbir Arif. 2018.Enabling Input on Tiny/Headless Systems Using Morse Code. https://www.asarif.com/pub/Gueorguieva_CCBM2018_Morse_Code.pdf

[10] R. L. Hershman and W. A. Hillix. 1965. Data Processing in Typing: Typing Rateas a Function of Kind of Material and Amount Exposed. Human Factors 7, 5 (Oct.1965), 483–492. https://doi.org/10.1177/001872086500700507

[11] Tomonari Kamba, Shawn A. Elson, Terry Harpold, Tim Stamper, and PiyawadeeSukaviriya. 1996. Using Small Screen Space More Efficiently. In Proceedings ofthe SIGCHI conference on Human factors in computing systems common ground- CHI ’96. ACM Press, Vancouver, British Columbia, Canada, 383–390. https://doi.org/10.1145/238386.238582

[12] Per Ola Kristensson and Keith Vertanen. 2014. The Inviscid Text Entry Rateand Its Application as a Grand Goal for Mobile Text Entry. In Proceedings ofthe 16th international conference on Human-computer interaction with mobiledevices & services (MobileHCI ’14). ACM, Toronto, ON, Canada, 335–338. https://doi.org/10.1145/2628363.2628405

[13] Reena Lee. 2016. Gboard, Now Available for Android. https://blog.google/products/search/gboard-now-on-android/ Library Catalog: blog.google.

[14] Frank Chun Yat Li, Richard T. Guy, Koji Yatani, and Khai N. Truong. 2011. The1Line Keyboard: A QWERTY Layout in a Single Line. In Proceedings of the 24thAnnual ACM Symposium on User Interface Software and Technology (UIST ’11).ACM, New York, NY, USA, 461–470. https://doi.org/10.1145/2047196.2047257

[15] Yunzhi Li, Yinan Xu, Xiang Li, Jie Liu, and Nianlong Li. 2018. TapFlick: CombiningTap and Flick for Text Entry on Touchscreen Devices. In Proceedings of the SixthInternational Symposium of Chinese CHI (ChineseCHI ’18). ACM, New York, NY,USA, 120–123. https://doi.org/10.1145/3202667.3202686 event-place: Montreal,QC, Canada.

[16] I. Scott MacKenzie and R.William Soukoreff. 2003. Phrase Sets for Evaluating TextEntry Techniques. In CHI ’03 Extended Abstracts on Human Factors in ComputingSystems (CHI EA ’03). ACM, New York, NY, USA, 754–755. https://doi.org/10.1145/765891.765971

[17] I. Scott Mackenzie, Shawn X. Zhang, and R. William Soukoreff. 1999. Text EntryUsing Soft Keyboards. Behaviour & Information Technology 18, 4 (Jan. 1999),235–244. https://doi.org/10.1080/014492999118995 Publisher: Taylor & Francis.

[18] Gulnar Rakhmetulla and Ahmed Sabbir Arif. 2020. Senorita: A Chorded Keyboardfor Sighted, Low Vision, and Blind Mobile Users. In Proceedings of the 2020 CHIConference on Human Factors in Computing Systems (CHI ’20). ACM, Honolulu,HI, USA, 1–13. https://doi.org/10.1145/3313831.3376576

[19] Marco Romano, Luca Paolino, Genoveffa Tortora, and Giuliana Vitiello. 2014.The Tap and Slide Keyboard: A New Interaction Method for Mobile Device TextEntry. International Journal of Human–Computer Interaction 30, 12 (Dec. 2014),935–945. https://doi.org/10.1080/10447318.2014.924349

[20] Yuan-Fu Shao, Masatoshi Chang-Ogimoto, Reinhard Pointner, Yu-Chih Lin, Chen-Ting Wu, and Mike Chen. 2016. SwipeKey: A Swipe-based Keyboard Designfor Smartwatches. In Proceedings of the 18th International Conference on Human-Computer Interaction with Mobile Devices and Services (MobileHCI ’16). ACM, NewYork, NY, USA, 60–71. https://doi.org/10.1145/2935334.2935336

[21] Ken C.F. Tan, Edwin Ng, and Julian J.S. Oh. 2003. Development of a Qwerty-TypeReduced Keyboard System for Mobile Computing: Tengo. Proceedings of theHuman Factors and Ergonomics Society Annual Meeting 47, 6 (Oct. 2003), 855–859.https://doi.org/10.1177/154193120304700603

[22] Stephen P. Witte and Roger D. Cherry. 1986. Writing Processes and WrittenProducts in Composition Research. In Studying Writing: Linguistic Approaches (1ed.). Sage Publications, Beverly Hills, CA, USA, 112–153.